Adrian,

The problem may be frame related, as opposed to RTP related. The code I looked at deals with frames, and I'm not sure where the audio data is delivered via RTP. I did some research today and determined that the skips/pops in the recordings correlate with dropped audio in the calls. This occurs regularly on our quad-processor box with a SINGLE call on it and one dynamic agent logged in. The problem does not seem as bad on the call, because sometimes the dropped audio is during a period of silence, but it can be heard as a pop on the recording. This is likely the result of garbage data at that point in the file, and the particular sound it produces probably varies from format to format. The format we are using is PCM. The calls themselves are u-Law.

I'm beginning to believe that the problem has something to do with the locking that occurs around the code I looked at, but I need to dig a little deeper to verify my hunch. Scroll down to see my analysis of the related channel.c code.

If anyone has experienced anything like I've been describing, please share your story with me. Are these drops just a fact of life with Asterisk or VoIP in general?

We have no hardware timing device on the box (no Zap hardware) and are using the 2.6 kernel as the timing source. Digium tech support told us this is better than ztdummy, which we were using before. We experienced the same problems then, as well. Could a lack of a hardware timing source be our problem?

Thanks,

Matthew Roth
InterMedia Marketing Solutions
Software Engineer and Systems Developer

Adrian Carter wrote:

Matt,
I have a similar issue to the 'Skips and Pops' with the On Hold music on my Ast 1.2.1 box. I've tried moving stuff to a RAM Disk, yet I still get reports from agents that callers report that the 'music on hold sounds horrible'.

It has squeaks and pops.. kind of like digital satelite distortion.. and looking at what you've said below, it kind of makes sense, since I can gurantee to re-create the problem by just dragging out a sip phone and dropping myself in a queue - All MoH sounds terrible, but the "Queue Announcements" (Your caller X in the queue) come out perfectly everytime.

The one thing that doesn't sit with your explanation : I seem to have ZapChannel users complain of the same problem. Im newish to Asterisk, but I thought RTP only came into play on SIP/IAX/MGCP calls ... So the fact that I seem to have the problem when calling from a CO Trunk line (well, Inbound PRI) into a digium 4 port PRI card means it couldn't be RTP related? just frame related?

Adrian

Matt Roth wrote:

List users,

I've traced the writing of the leg files to two functions in channel.c:

ast_write()
ast_read()

They both contain similar code, so I'm going to limit my analysis to one of them. If I'm misunderstanding anything or am flat out wrong, please don't hesitate to correct me. Your input is what I'm looking for.

Below is a code fragment, with some extraneous stuff removed for brevity and some comments describing what I believe is happening. Please take a look at it, paying special attention to the comments I've added.

My understanding is that if the channel is locked, the function will wait on the ast_mutex_lock() call until it is unlocked. Once it is unlocked, the function attempts to compensate for any loss of leg file synchronization by jumping the file pointer forward by some value based on the number of dropped frames. This puts a gap in the leg file which manifests itself as a popping sound in the format we are using (PCM), but which probably sounds a little different in other formats.

My main concern is if this fragment is responsible for writing the frame to its destination via RTP. If it is, the skips and pops in the recordings would likely manifest themselves as dropped audio on the calls. Is this correct? Do problems in the recordings indicate parallel problems in call quality? The reports from our agents don't seem to support this, but looking at the code worries me.

Thank you very much,

Matthew Roth
InterMedia Marketing Solutions
Software Engineer and Systems Developer

=====================================================================

int ast_write(struct ast_channel *chan, struct ast_frame *fr)
{
   /* Lock the channel */
   ast_mutex_lock(&chan->lock);

  /*
     ... Code Omitted ...
                          */

  /* Check to see if the channel is blocking */
   CHECK_BLOCKING(chan);

  /* Switch based on frame type */
   switch(fr->frametype) {

  /*
     ... Code Omitted ...
                          */

  /* Handle voice frames */
   default:
       /* Validate a function pointer */
       if (chan->pvt->write) {
           /* Validate another function pointer */
           if (chan->pvt->writetrans) {
               f = ast_translate(chan->pvt->writetrans, fr, 0);
           } else
               f = fr;
           /* Validate the frame pointer */
           if (f) {
/* I'm not sure what this does, so please let me know if you do. I really hope that it's not responsible for writing the frame out to its destination via RTP. */
              res = chan->pvt->write(chan, f);
/* If this channel is being monitored write the frame to the appropriate leg file */
               if( chan->monitor &&
                       chan->monitor->write_stream &&
                       f && ( f->frametype == AST_FRAME_VOICE ) ) {
                            /*
If some frames have been missed, jump the leg file pointer forward to keep the leg files synchronized.

!!! I BELIEVE THIS IS THE SOURCE OF THE SKIPS AND POPS IN THE RECORDINGS !!! */
              #ifndef MONITOR_CONSTANT_DELAY
int jump = chan->insmpl - chan->outsmpl - 2 * f->samples;
                   if (jump >= 0) {
if (ast_seekstream(chan->monitor->write_stream, jump + f->samples, SEEK_FORCECUR) == -1) ast_log(LOG_WARNING, "Failed to perform seek in monitoring write stream, synchronization between the files may be broken\n");
                       chan->outsmpl += jump + 2 * f->samples;
                   } else
                       chan->outsmpl += f->samples;
#else
                   int jump = chan->insmpl - chan->outsmpl;
                   if (jump - MONITOR_DELAY >= 0) {
if (ast_seekstream(chan->monitor->write_stream, jump - f->samples, SEEK_FORCECUR) == -1) ast_log(LOG_WARNING, "Failed to perform seek in monitoring write stream, synchronization between the files may be broken\n");
                       chan->outsmpl += jump;
                   } else
                       chan->outsmpl += f->samples;
#endif
               /* Write the frame to the leg file */
               if (ast_writestream(chan->monitor->write_stream, f) < 0)
ast_log(LOG_WARNING, "Failed to write data to channel monitor write stream\n");
               }
           } else
               res = 0;
       }
   }

  /*
     ... Code Omitted ...
                          */

  /* Set the channel as not blocking */
   chan->blocking = 0;

  /*
     ... Code Omitted ...
                          */

  /* Unlock the channel and return */
   ast_mutex_unlock(&chan->lock);
   return res;
}

=====================================================================
_______________________________________________
--Bandwidth and Colocation provided by Easynews.com --

Asterisk-Users mailing list
To UNSUBSCRIBE or update options visit:
  http://lists.digium.com/mailman/listinfo/asterisk-users


_______________________________________________
--Bandwidth and Colocation provided by Easynews.com --

Asterisk-Users mailing list
To UNSUBSCRIBE or update options visit:
  http://lists.digium.com/mailman/listinfo/asterisk-users

Reply via email to